an optimal strategy of natural hedging ... - longevity risk
TRANSCRIPT
1
An Optimal Strategy of Natural Hedging for
a General Portfolio of Insurance Companies
Hong-Chih Huang1 Chou-Wen Wang2 De-Chuan Hong3
ABSTRACT
With the improvement of medical and hygienic techniques , life insurers may gain a
profit whilst annuity insurers may suffer losses because of longevity risk. In this paper,
we investigate the natural hedging strategy and attempt to find an optimal collocation
of insurance products to deal with longevity risks for life insurance companies.
Different from previous literatures, we use the experienced mortality rates from life
insurance companies rather than population mortality rates. This experienced
mortality data set includes more than 50,000,000 policies which are collected from the
incidence data of the whole Taiwan life insurance companies. On the basis of the
experienced mortality rates, we demonstrate that the proposed model can lead to an
optimal collocation of insurance products and effectively apply the natural hedging
strategy to a more general portfolio for life insurance companies.
Keyword: Longevity risk, Natural hedging strategy, Experienced mortality rates
1 Professor, Department of Risk Management and Insurance, National Chengchi University, Taipei, Taiwan. E-mail: [email protected] *Corresponding author. 2 Associate Professor, Department of Risk Management and Insurance, National Kaohsiung First University of Science and Technology, Kaohsiung, Taiwan. 3 Department of Risk Management and Insurance, National Chengchi University, Taipei, Taiwan.
2
1.Introduction
According to the publication-Sigma of Swiss Re, the life expectancy of human
around the world will increase 0.2 year per year. Because the mortality rate has
improved rapidly for the past decades, longevity risk has become an important topic.
In the past two decades, a wide range of mortality models have been proposed and
discussed (Lee-Carter, 1992; Brouhns et al., 2002; Renshaw and Haberman, 2003;
Koissi et al., 2006; Melnikov and Romaniuk, 2006; Cairn, Blake and Dowd, 2006).
Among them, the Lee-Carter (LC) (1992) model is probably the most popular choice,
because it is easy to implement and provides acceptable prediction errors.
Constructing delicate mortality model for the use of pricing is one solution to hedge
longevity risk for both life insurance and annuity products. However, this solution is
often difficult to apply into practice because of market competition. Even though
insurance companies have ability to build a delicate mortality model to catch the
actual future mortality improvement, they may not be able to price and sell annuity
products using the mortality rate derived from this mortality model since it might be
too expensive to sell these annuity products with market competition. Another
possible solution to hedging longevity risk is to use the mortality derivatives, such as
Survival Bonds and Survival Swaps. Blake and Burrows (2001) propose the concept
of Survival Bond and insurance companies can hedge longevity risk based on it. Cairn,
Blake and Dowd (2006) propose Survival Swaps, which is a contract to exchange
cash flows in the future based on the survivor indices. Although mortality derivatives
are easy and convenient to use but there are still many obstacles in mortality
derivative. The special purpose vehicles must pay more attention on their customers
and counterparty, and that means insurance companies have to pay a huge amount of
transaction cost on mortality derivatives. Another solution is natural hedging.
3
Insurance companies can optimize the collocation of its products, annuities and life
insurances, to hedge longevity risk. This approach can be done internally in an
insurance company. Therefore it is more convenient and practical for insurance
company to hedge longevity risk by using this method.
Natural hedging is a relatively new topic in actuarial field, so few papers have
studied this issue. Wang, Yang and Pan (2003) investigate the influence of the changes
of mortality factors and propose an immunization model to hedge mortality risks. Cox
and Lin (2007) indicate that natural hedging utilizes the interaction of life insurance
and annuities to a change in mortality to stabilize aggregate cash outflows. And they
drew a conclusion that natural hedging is feasible and mortality swaps make it
available widely. Wang, Huang, Yang and Tsai (2010) analyze the immunization
model mentioned above and use effective duration and convexity to find the optimal
product mix for hedging longevity risk. However, their paper uses the same mortality
rate for both life insurance and annuity products due to lack of experience mortality
data. This is definitely not true in practice.
Different from those previous literatures, we consider the “variance” effect related
to both uncertainties of mortality rate and interest rate and the “miss-pricing” effect
induced by the mortality improvement in constructing natural hedging portfolio. In
addition, we use the experienced mortality rates from life insurance companies rather
than population mortality rates. This experienced mortality data set includes more
than 50,000,000 policies which are collected from the incidence data of the whole
Taiwan life insurance companies. Because we don’t have real annuity mortality data,
we regard the experience mortality rate of life insurance policies with heavy principal
repayment as annuity mortality rate. Both experience mortality rates between with and
without principal repayment are correlated-but not perfectly negative correlated.
4
Therefore, it is not possible to perfectly hedge longevity risk as Wang, Huang, Yang
and Tsai (2010) when we consider both life and annuity mortality rates. Besides, we
consider the pricing differences in those insurance products between period-mortality
basis and cohort-mortality basis. Our objective is to minimize the variation of the
change of total portfolio’s value and the differences between different pricing bases.
On the basis of these experienced mortality rates, the proposed model in this paper
appropriately provides an optimal collocation of insurance products and effectively
applies the natural hedging strategy to a more general portfolio for life insurance
companies.
The reminder of this paper is organized as follows. In Section 2, we review the
mortality model and interest rate model, proposing our portfolio model with
“variance” effect and “miss-pricing” effect. In Section 3, we calibrate the Lee-Carter
model by using the experienced mortality rates from Taiwan Insurance Institute (TII).
Section 4 contains the numerical analysis of our model. Section 5 summarizes the
paper and gives conclusions and suggestions.
2.Model Setting
2.1. Mortality Model: Lee-Carter Model
Lee and Carter proposed this mortality model in 1992. They found out the fitting
and forecasting ability of this model is superior and simple but powerful
one-parameter model. Since then, Lee-Carter model has been one of the most popular
stochastic mortality models. Their model is as follows:
, ,ln( )x t x x t x tm k , (1)
where
5
,x tm : The central death rate for a person aged x at time t.
x : The average age-specific mortality factor.
x : The age-specific improving factor.
tk : The time-varying effect index.
In their study, the time effect index tk can be estimated by using an ARIMA (0,1,0)
process. Note that the continuous-time limit of the ARIMA (0,1,0) process of tk can
be expressed as ( )t k k kdk u dt dZ t , where 0
( )T
k tZ t
is a standard Brownian
motion.
There are lots of papers investigating the fitting of Lee-Carter parameters and two
most popular methods are SVD (Singular Value Decomposition) and Approximation.
In their paper, Lee and Carter use SVD to find the parameters. Wilmoth (1993)
propose a modified Approximation method, Weighted Least Squares, to avoid the
zero-cell problem. We use Approximation method to estimate the parameters of
Lee-Carter model because missing values problem did exist in our data. The
computational steps can be written as follows:
1.The fitted values of x equal the average of ,ln( )x tm at age x over time.
2.Employ the standard normalizing constraints on both x and tk , such that
1xx
b and 0tt
k .
3. tk is equal to the sum of ,ln( )x t xm over each age.
4.Using the regression model without intercept but with dependent variables tk and
,ln( )x t xm , we can obtain the coefficient x .
6
Following the above steps, we can obtain the estimated parameters ,x x and tk
to forecast the future mortality rates.
2.2. Interest Rate Model: CIR Model
Cox, Ingersoll and Ross (1985) specifies that the instantaneous interest rate
follows a square root process, also named the CIR process as follows:
( ) ( )t t r t rdr a b r dt r dZ t , (2)
where a denotes the speed of mean-reverting adjustment; b represents the long term
mean of interest rates; r is interest rate volatility; and 0
( )T
r tZ t
is standard
Brownian motion modeling the random market risk factor. In CIR model, the interest
rate towards the long run level b with speed of adjustment governed by the strictly
positive parameter a. in this paper, we assume that 22 rab ; therefore, an interest
rate of zero is also precluded.
We can transform the continuous-time process into a discrete-time version as
follows:
1 1 1( )t t t r t tr r a b r r , (3)
where t is a standard normal random variable. We can use Equation (3) to simulate
the future interest rates with the setting origin 0r . The long-term interest rate can be
estimated by term structure of interest rate. According to CIR model, we can calculate
the price of one unit zero-coupon bond at time t with maturity date t+T as follows:
( , )( , , ) ( , ) tB t T rtP r t T A t T e , (4)
where
7
22 /[( ) ]/22( , )
( )( 1) 2
raba T
T
eA t T
a e
, (5)
2( 1)( , )
( )( 1) 2
T
T
eB t T
a e
, (6)
2 2( ) 2 ra , (7)
2.3. General Portfolio Model
In our model, we adopt Lee-Carter mortality model and CIR Interest rate model.
The insurance company’s portfolio contains zero-coupon bonds, annuities and life
insurances with different ages and genders. The factors affecting the total value of this
portfolio are mortality rate and interest rate. The model is as follows:
, ,
, 1,2
( ) ( ), , ,s g s g Bx x t B t
s A L g x i
V t N V m t r t N V r t
, (8)
where V represents the value of the insurance portfolio; ( , )BtV r t is the value of
one unit of zero-coupon bond with fact value equal to one;
,( )x x t tm t m ; ,1( ( ), , )Ax tV m t r t ( ,2( ( ), , )A
x tV m t r t ) denotes the value of one unit of
annuity policy issued to the cohort that consists of males (females) aged x at time 0;
,1( ( ), , )Lx tV m t r t ( ,1( ( ), , )L
x tV m t r t ) denotes the value of one unit of life insurance policy
issued to the males (females) that consists of females aged x at time 0; BN
represents the number of units invested in zero-coupon bonds; and ,A gxN ( ,L g
xN )
denotes the number of units allocated in the annuity (life insurance) policies.
From Lee-Carter model, we can get the following information ,x x and tk by
using approximation method and the estimated force of mortality is as follows:
8
,ln( )x t t x t x t tm k . (9)
Because the continuous-time limit of the ARIMA (0,1,0) process of tk can be
expressed as ( )t k k kdk u dt dZ t , differencing Equation (9) yields the continuous
-time representation of log mortality rate as follows:
ln( ) ( )x t x t k k kd m u dt dZ t , (10)
or equivalently,
, ,0
, , , , , , ,ln( ) ln( ) ( ) , s or , 1 2x t t x t
s g s g s g s g s g s g s gx t k x t k km m u t Z t L A g or
. (11)
By using the Ito’s lemma, we can transform the dynamic of logarithm of mortality
into the dynamic of mortality rate as follows:
,2ln( )
, , , , ,
2 2, ,
1ln ln
21
= ( )2
x t tmx t t x t t x t t x t t x t t
x t t x t k x t k k x t t x t k
d m d e m d m m d m
m u dt dZ t dt m dt
2 2, , ,
1 = ( )
2x t t x t k x t t x t k x t t x t k km u m dt m dZ t
, (12)
or equivalently,
2 2, , , , , , , , , , ,+ + +
1( ) = ( ) ( ) ( ) ( )
2s g s g s g s g s g s g s g s g s g s g s gx x x t k x x t k x x t k kd m t m t u m t dt m t dZ t
, 1 2for s L or A g or , (13)
In our model, to capture the covariance structure of mortality rates, we assume that
the four mortality risk factors, ,2AkZ , ,1A
kZ , ,2LkZ and ,1L
kZ are dependent standard
Brownian motions. For the tractability of our analysis, we decompose them into the
linear combination of four independent standard Brownian motions
0
( )T
ki ki tZ Z t
, =1,2,3,4i , by using Chelosky decomposition as follows:
9
,21
,121 22 2
,231 32 33 3
,141 42 43 44 4
1 0 0 0 ( )( )
0 0 ( )( )
0 ( )( )
( )( )
Akk
Akk
Lkk
Lkk
dZ tdZ t
a a dZ tdZ t
a a a dZ tdZ t
a a a a dZ tdZ t
. (14)
Utilizing the Ito’s lemma, we investigate the change of total insurance portfolio
value with respect to the change of mortality rate and interest rate as follows:
, 2 , 2
, , , , 2 2
, , 2 2, 1,2 , 1,2
1 1( ) ( ( )) ( )
2 2( ) ( )
s g s g
s g s g s g s gx x
x x x xs g s gs L A g x i s L A g x ix x
t t
V VV VN dm dr N dm t dr
m r m rdV t t
4
0 51
( ) ( )t it ki t ri
Q dt Q dZ t Q dZ t
, (15)
where 0 1 2 3 4, , , ,t t t t tQ Q Q Q Q and 5tQ are defined as follows:
,, , , , , , 2 , 2
0 +,, 1,2
1( ( ) ( ) )
2
s gs g s g s g s g s g s g s gx
t x x x t k x x t ks gs L A g x i x
VQ N m t m t
m
2 , 2, , , , 2 2
, 2 2
1 1+ ( ( ) ) ( )
2 2
s gs g s g s g s gxx x x t k rs g
x
V V VN m t a b r r
m r r
, (16)
,2 ,1,2 ,2 ,2 ,2 ,1 ,1 ,1 ,1
1 + 21 +,2 ,1( ) ( )
A AA A A A A A A Ax x
t x x x t k x x x t kA Ax i x ix x
V VQ N m t a N m t
m m
,2 ,1
,2 ,2 ,2 ,2 ,1 ,1 ,1 ,131 + 41 +,2 ,1
( ) ( )L L
L L L L L L L Lx xx x x t k x x x t kL L
x i x ix x
V Va N m t a N m t
m m
, (17)
,1 ,2,1 ,1 ,1 ,1 ,2 ,2 ,2 ,2
2 22 + 32,1 ,2( ) ( )
A LA A A A L L L Lx x
t x x x t k x x x t kA Lx i x ix x
V VQ a N m t a N m t
m m
,1
,1 ,1 ,1 ,142 +,1
( )L
L L L Lxx x x t kL
x i x
Va N m t
m
, (18)
,2 ,1,2 ,2 ,2 ,2 ,1 ,1 ,1 ,1
3 33 43,2 ,1( ) ( )
L LL L L L L L L Lx x
t x x x t k x x x t kL Lx i x ix x
V VQ a N m t a N m t
m m
, (19)
,1,1 ,1 ,1 ,1
4 43 ,1( )
LL L L Lx
t x x x t kLx i x
VQ a N m t
m
, (20)
10
5 Q t r t
Vr
r
; (21)
the V r and 2 2V r are of the form:
,,
, 1,2
s g Bs g xx B
s L A g x i
VV VN N
r r r
, (22)
2 ,2 2,
2 2 2, 1,2
s g Bs g xx B
s L A g x i
VV VN N
r r r
. (23)
Under the assumption that mortality rates and financial risk are independent, the
terms , 1,2,3,4ki rdZ dZ i are zero in our model. We apply the concepts of effective
durations and convexities to estimate the first-order and second-order derivatives in
our model as follows:
( , ) ( , )
2 ( , )
V V m r V m r
m V m r m
, (24)
( , ) ( , )
2 ( , )
V V m r V m r
r V m r r
, (25)
2
2 2
( , ) ( , ) 2 ( , )
( , )( )
V V m r V m r V m r
m V m r m
, (26)
2
2 2
( , ) ( , ) 2 ( , )
( , )( )
V V m r V m r V m r
r V m r r
, (27)
where (1 ); (1 );m m m m (1 ); (1 );r r r r represents a
strictly positive change rate of mortality rate and interest rate; and m and r
satisfy the following expression:
1x x
m m m , (28)
1t t
r r r . (29)
11
By virtue of Equation (15), the variance of the change of total insurance portfolio
value is as follows:
52
=1
( ( )) ( )j tj
Var dV t Q , (30)
which is determined by the parameters of Lee-Carter model and CIR model as well as
the first-order and second-order derivatives defined in Equations (24)-(27).
Insurers minimize the variance of portfolio returns with respect to the change in
mortality rate and interest rate because their risk profiles are mainly consist of the
longevity risk and interest rate risk. Therefore, we incorporate the variance in
Equation (30), named as “variance effect”, in the objective function for optimally
allocating the insurance portfolio with annuity policies and life insurance policies.
Besides, we consider different pricing basis between period-mortality basis and
cohort-mortality basis. The mortality rate without the improvement effect is called
period mortality rate which lots of insurance companies apply to price insurance
policies. The mortality rate with improvement effect, called cohort mortality rate,
consists with the tendency of future mortality rate. Consequently, the difference of
portfolio value evaluated between these two bases can be regarded as a miss-pricing
error. The pricing difference of this portfolio is of the form:
, , ,
, 1,2
( ( ), , ) ( ( ), , )s g s g s gx period x t cohort x t
s A L g x i
D N V m t r t V m t r t
, (31)
where ,( ( ), , )s gperiod x tV m t r t denotes the policy value with period-mortality basis whilst
,( ( ), , )s gcohort x tV m t r t denotes the policy value with cohort-mortality basis. As a result,
we also incorporate the pricing difference, named as “miss-pricing effect”, in the
objective function.
12
The purpose of our paper is to find a feasible policy collocation in the risk profile
of insurance companies to minimize the variance effect with a weight (1-θ) and the
miss-pricing effect with a weight θ. If the insurance companies put more emphasis on
the variance effect, they tend to control the change in the insurance portfolio due to
the unexpected shocks in mortality rate and interest rate. If they put more emphasis on
the miss-pricing effect, they aim to minimize the miss-pricing error by increasing the
weight θ. Our objective function can be expressed as follows:
,
5, 2 2
,=1
( , ) min (1 ) ( )s gx B
s gx B j t
N Nj
f N N Q Dq q= - +å . (32)
3.Mortality Data
The mortality data were collected via Taiwan Insurance Institute (TII) and include
more than 50,000,000 policies of life insurance companies in Taiwan. The original
data are categorized by ages, gender and sorts. We use the original data to construct
four Lee-Carter mortality tables, including female annuity (fa), male annuity (ma),
female life insurance (fl) and male life insurance (ml). The maximal age of the
original data is about 85. The parameters of LC model are calibrated by using
approximation method and are dipicted in Figure 1.
[Insert Figure 1 here]
In order to capture longevity risk, we utilize extrapolation to extend maximal age
from 85 to 100. For the purpose of forecasting the future mortality rate, in Table 1 we
calculate the standard deviation ( k ) of fat , ma
t , flt and ml
t .
[Insert Table 1 here
13
We then use Cholesky decomposition method to transform the dependent random
variables into a linear combination of independent random variables. First, we
compute the correlation matrix (M) of fat , ma
t , flt and ml
t as follows:
1 0.774017 -0.09217 0.243753
0.774017 1 -0.1304 0.358255
-0.09217 -0.1304 1 0.455803
0.243753 0.358255 0.455803 1
M
. (33)
Based on the Cholesky decomposition, the lower triangular matrix (R), which satisfies
TM R R , can be expressed as follows:
1 0 0 0
0.7740 0.6332 0 0
0.0922 0.0933 0.9914 0
0.2438 0.2678 0.5076 0.7818
R
. (34)
Therefore, based on the lower triangular matrix (R), Equation (14) can be rewritten as
follows:
,21
,12
,23
,14
( )1 0 0 0( )
( )0.7740 0.6332 0 0( )
( )0.0922 0.0933 0.9914 0( )
( )0.2438 0.2678 0.5076 0.7818( )
Akk
Akk
Lkk
Lkk
dZ tdZ t
dZ tdZ t
dZ tdZ t
dZ tdZ t
. (35)
4.Numerical Analysis
4.1Scenario 1 : θ = 0 (variance effect)
We first assume that the weight θ is zero, and then we can observe the effect caused
by the variance effect only. Whith loss of generity, in the squeal we assume that
a=0.1663, b=0.0606 and r =4.733% for the parameters of CIR model. We begin
14
with a simple case, in which there are only two policies in the portfolio: an annuity
and a life insurance policy. We can capture the corresponding hedging relation from
those simple cases. Given one unit value of annuity due, we can find the optimal
value (unit) of life insurance. Assuming that (the unexpected change rate of future
mortality rate and interest rate) are 5%, 10%, and 15%, Table 2 represents the optimal
units of life insurance to hedge one unit value of annuity due for different unexpected
change rates.
We see from Table 2 that the unexpected change rate of mortality and interest rate
doesn’t have obvious impact on the optimal units held in life insurance. The reason is
that the effective durations and convexities of each product are almost the same. In
Table 3 we take the life insurance policy of female with age 30 (fl30) as an example.
The variance increases only slightly with larger unexpected change rate because of the
slightly change of duration and convexity. We use 10% unexpected change rate of
both mortality and interest rate for the following scenario analyses.
[Insert Table 2 here]
[Insert Table 3 here]
We calculate the corresponding optimal units of different life insurances to hedge
one unit of female annuity due. The results are shown in Table 4. From Table 4, we
find that we are not able to reduce the total variance by holding female life insurances
to hedge annuity policy, but we can make it through male life insurances. Thus, Table
4 shows that the variance of the portfolio of female annuity and male life insurance
are smaller than that of the portfolio of female annuity and female life insurance.
[Insert Table 4 here]
15
We further investigate hold one unit value of male annuity, and the following the
optimal units of life insurance with different genders and ages. Table 5 shows the
corresponding optimal units of different life insurances to hedge one unit of male
annuity due. We find that results are similar to those observed from Table 4. In
addition, comparing the right-hand side of Table 4 with that of Table 5, we observe
that the optimal units of male life insurance to hedge one unit of male annuity60 (aged
60, male annuity) is larger than those of female annuity60.
[Insert Table 5 here]
We expect to hedge longevity risk of annuity product by holding some units of life
insurances because we know that the sign of effective duration of annuity and life
insurance are opposite. We can’t obtain the hedging effect from holding life insurance
because the coefficient of Cholesky decomposition of female life insurance is
negative. However, the coefficient of male life insurance is positive, so it is possible
to minimize the variance effect by holding some male life insurance policies. The
reason that we must hold more male life insurance to hedge male annuity60 than
female annuity60 is that the magnitude of male annuity’s duration and mortality rate
are larger than those of female, as shown in Table 6. Consequently, we must hold
more units of male life insurances to hedge 1 unit of male annuity60.
[Insert Table 6 here]
We further want to compare the differences of optimal hedging strategies between
holding one unit of annuity due and one unit of deferred annuity. The comparison
results are as presented in Tables 7 and 8. From Table 7-1 & 7-2 and Table 8-1 & 8-2,
for deferred annuities, we need to hold less life insurance policy to hedge the
mortality uncertainty. In addition, we observe that, with the increase of insured’s ages,
16
we need less life insurance units to hedge the corresponding annuity. Although the
duration of younger life insurances is larger than that of elder, the elder’s mortality
rate is much higher than the younger’s mortality rate. Therefore, it is more possible
for the insurance company to suffer an instant claim by the elder insured. So the
insurance company needs less units of the elder life insurance to offset the longevity
risk of annuities.
[Insert Table 7 here]
[Insert Table 8 here]
4.2Scenario 2: θ=1 (miss-pricing effect )
In this section we ignore the variance effect and discuss the miss-pricing effect.
Without consideration of the variance effect in the objective function, we can find a
perfect-hedge unit of life insurance product to hedge one unit of annuity, and vice
versa. Therefore, the value of objective function is always zero in the optimal
situation. In the following analysis, we just show the optimal units of life insurance
polices, corresponding to fixedly holding one unit of annuity policy. In Figure 2, we
depict the levels of under-pricing for annuity policies and over-pricing for life
insurance policies under different ages. We find that, for annuity products,
under-pricing problem of male is more serious than that of female. For life insurance
products, under-pricing problem of female is more serious than that of male. In
addition, we observe that the magnitude of miss-pricing problem decreases as the
issued age increases.
[Insert Figure 2 here]
17
According to Table 9 and Table 10, unlike the result in Scenario 1, we are able to
hedge longevity risk of annuity products through female life insurance products in
Scenario 2. However, with the issued age increases, the holding tendency goes up.
This outcome is totally different from the results in Scenario 1. This implies that
insurance companies need to take very different hedging strategies to reduce variance
effect and miss-pricing effect respectively.
[Insert Table 9 here]
[Insert Table 10 here]
4.3Scenario 3: 0 < θ <1
In practice, insurance companies should take account of both variance and
miss-pricing effects simultaneously to hedge longevity risk. This means the value of θ
should between 0 and 1 but not equal to 0 or 1. In this section, we discuss optimal
hedging strategies by varying the level of θ. We first consider holding one unit of
annuity due and find the optimal units of life insurance. The results of optimal units of
female life insurance are expressed in Table 11.
[Insert Table 11 here]
From Table 11, we observe that the optimal units of female life insurance decrease
as the weight θ decreases. That is, when we put less emphasis on miss-pricing effect,
we tend to minimize the variation of the change of total portfolio value. As a result,
the less unit of life insurance we hold, the smaller the objective function is. According
to Table 11, the patterns of optimal units of female life insurance to hedge one unit of
annuity60 for both male and female under different value of θ are similar, but the
magnitude of ma60 is larger than that of female annuity60, which is consistent with as
the results in Scenarios 1 & 2 mentioned above. In addition, we find that the female
life insurances are unable to reduce the variance of total portfolio value in Scenario 1,
18
which explains the reason why we need less units of life insurance as the weight θ
decreases.
In Table 11, insurance companies shall hold smallest optimal units as θ = 0 and
hold the largest optimal units as θ =1. When θ is equal to 0.1, 0.01 and 0.005
respectively, the patterns of optimal units are similar to the case with θ =1. In the case
of θ = 0.001, we find the pattern of optimal units is similar to the case of θ =1 in
younger age, but in older ages the pattern of optimal units will decrease like the case
of θ = 0. In Scenario 1, we observe that the variance effect is stronger in older age
because the variance effect is positively related to the mortality rate and effective
duration. So we will observe the interaction of both variance and difference effect in
the case of θ = 0.001 in the following analysis.
In Table 12, we investigate optimal units of male life insurance to hedge one unit of
annuity60 for both male and female under different value of θ, and the results are
shown as follows. Compared the results in Table 12 with those in Table 11, the
optimal units of male life insurance do not always decrease as the weight θ decreases.
In Scenario 1, we can utilize male life insurances to reduce total variance term,
however, we also take miss-pricing effect into consideration at the same time.
Therefore, the outcomes in Table 11 & 12 are the interaction between these two
effects.
[Insert Table 12 here]
According to the results of Table 11 and 12, we choose the weight θ=0.001 as an
example because there exists an interaction between two effects in average. The
results of optimal units of male and female life insurance to hedge one unit of annuity
for different ages and genders are expressed in Figure 3. Figure 3 shows that the
patterns of optimal units of male and female life insurance to hedge one unit of
19
annuity are similar for different ages and genders. We see from Figure 3 that the trend
of optimal units among different ages and genders are similar. The main differences
among them are the magnitude of optimal units. In general, optimal units of male life
insurance to hedge one unit of annuity is slightly larger than those of female life
insurance. Optimal units of male or female life insurance to hedge a younger age of
annuity are larger than those to hedge an older age of annuity. In addition, optimal
units of male or female life insurance to hedge one unit of male annuity are larger
than those to hedge a female annuity.
[Insert Figure 3 here]
5.Conclusion and Suggestion
In this paper, we propose a natural hedging model, which can take account of two
important effects of longevity risk at the same time. The first one is the variance of the
change of total portfolio value, and the second one is the miss-pricing effect. We can
hedge variations of the future mortality rate by the first effect, and hedge the present
miss-pricing by the second one.
Previous researches on natural hedging only take account of variance effect, but we
incorporate both variance and miss-pricing effects into our model in this paper. In
practice, lots of insurance companies indeed have a miss-pricing problem due to
mortality improvement over time. Therefore, we contribute to the existing literature
by showing that it is not suitable to ignore miss-pricing effect to hedge longevity risk.
Unlike the previous literatures, we use the experienced mortality rates from life
insurance companies instead of using population mortality rates. These differences
make our model more general and easier to apply into practice.
20
Using the experienced mortality rates, we separate the mortality rate from gender
and the type of insurance products, life insurance policies and annuity policies. We
use the correlations between these four types of mortality rates to hedge the variations
of the future mortality rate. Furthermore, we use Lee-Carter model to forecast the
future mortality rate and calculate the level of miss-pricing in practice. Then,
insurance companies can decide the relative significance of the variance and
miss-pricing effects by a weight θ. So we can effectively apply the natural hedging
strategy to a more general portfolio of life insurance companies. When we consider
the variance effect only, the optimal units of life insurance is affected by the effective
duration and mortality rate mostly. As a result, the optimal units of life insurance
decrease as the age of life insurer increases. However, when we consider the
miss-pricing effect only, the optimal units are totally decided by the period-cohort
difference of each product. The optimal collocation strategy decided by considering
the variance effect comes to an opposite conclusion by considering the variance effect
case. In this paper, we obtain optimal collocation solutions for both effects to hedge
longevity risk. Furthermore, the model proposed in this paper is easy to extend to a
general portfolio with realistic combinations of annuity and life insurance policies.
21
6.References
Blake, D., and Burrows, W. (2001), “Survivor Bonds: Helping to Hedge Mortality
Risk”, The Journal of Risk and Insurance 68: 339-348.
Brouhns, N., Denuit, M., and Vermunt, J. K. (2002), “A Poisson Log-Bilinear
Regression Approach to the Construction of Projected Life-Tables”,
Mathematics and Economics, 31: 373-393.
Cairns, A. J. G., Blake, D., and Dowd, K. (2006), “A Two-Factor Model for Stochastic
Mortality with Parameter Uncertainty: Theory and Calibration”, The Journal of Risk
and Insurance, 73: 687-718.
Cox, J. C., Ingersoll, Jr., J. E., Ross, S. A. (1985), “A Theory of the Term Structure of
Interest Rates”, Econometrica, 53: 385-408.
Cox, S. H., and Lin, Y. (2007), “Natural Hedging of Life and Annuity Mortality
Risks”, North American Actuarial Journa, 11(3): 1-15.
Dowd, K., Blake, D., Cairns, A. J. G., and Dawson, P. (2006), “Survivor Swaps”, The
Journal of Risk & Insurance, 73: 1-17.
Koissi, M. C., Shapiro, A. F., and Högnäs, G. (2006), “Evaluating and Extending the
Lee–Carter Model for Mortality Forecasting: Bootstrap Confidence Interval”,
Mathematics and Economics, 38: 1-20.
Lee, R. D., and Carter, L. R. (1992), “Modeling and Forecasting U.S. Mortality”,
Journal of the American Statistical Association, 87(419): 659-675.
Melnikov, A., and Romaniuk, Y. (2006), “Evaluating the Performance of Gompertz,
Makeham and Lee–Carter Mortality Models for Risk Management with Unit-Linked
Contracts”, Mathematics and Economics, 39: 310-329.
Formatted: Pattern: Clear(White)
Formatted: Pattern: Clear(White)
Formatted: Pattern: Clear(White)
22
Renshaw, A. E., and Haberman, S. (2003) “ Lee-Carter Mortality Forecasting with
Age Specific Enhancement”, Mathematics and Economics, 33: 255-272.
Wang, J. L., Huang, H.C., Yang, S. S., and Tsai, J. T. (2010), “An Optimal Product
Mix For Hedging Longevity Risk in Life Insurance Companies: The Immunization
Theory Approach”, The Journal of Risk and Insurance, 77: 473-497.
Wang, J. L., Yang, L. Y., and Pan, Y. C. (2003), “Hedging Longevity Risk in Life
Insurance Companies”, In Asia-Pacific Risk and Insurance Association, 2003
Annual Meeting.
Wilmoth, J. R. (1993), “Computational Methods for Fitting and Extrapolating the
Lee-Carter Method of Mortality Change”, Technical Report, Department of
Demography, University of California, Berkeley.
23
TABLE 1
Standard deviation of tk in four different mortality groups
fat ma
t flt ml
t
Std. 9.34057 10.75928 9.89995 8.26466
TABLE 2
Optimal units of life insurances with different (θ = 0)
1 unit fa60 1 unit ma60
fl30 ml30 fl30 ml30
N 4.51×10-5 0.1353 N 4.51×10-5 0.5411 =5%
f(N) 1.65×10-7 1.56×10-7 =5%
f(N) 1.23×10-6 1.07×10-6
N 4.51×10-5 0.1353 N 4.51×10-5 0.5411 =10%
f(N) 1.66×10-7 1.56×10-7 =10%
f(N) 1.23×10-6 1.07×10-6
N 4.51×10-5 0.1353 N 4.51×10-5 0.5412 =15%
f(N) 1.66×10-7 1.56×10-7 =15%
f(N) 1.23×10-6 1.07×10-6
TABLE 3
Effective duration (dur) and convexity (conv) with different (5%, 10%, 15%)
fl30 =5% =10% =15%
dur(m) 11.4104 11.4160 11.4255
dur(r) -49.5854 -50.2119 -51.2687
conv(mm) -133.783 -133.897 -134.087
conv(rr) 2,701.095 2,719.847 2,751.366
conv(mr) -348.594 -349.216 -350.201
24
TABLE 4
Optimal units of life insurances, given 1 unit of female annuity (θ = 0)
1 unit fa60 1 unit fa60 N f(N) N f(N)
fl30 4.51×10-5 1.66×10-7 ml30 0.1226 1.56×10-7 fl40 4.51×10-5 1.66×10-7 ml40 0.0863 1.62×10-7 fl50 4.51×10-5 1.66×10-7 ml50 0.0407 1.62×10-7 fl60 4.51×10-5 1.66×10-7 ml60 0.0355 1.62×10-7
TABLE 5
Optimal units of life insurances, given 1 unit of male annuity (θ = 0)
1 unit ma60 1 unit ma60
N f(N) N f(N)
fl30 4.51×10-5 4.88×10-7 ml30 0.5389 1.06×10-6
fl40 4.51×10-5 1.22×10-6 ml40 0.3887 1.06×10-6
fl50 4.51×10-5 1.22×10-6 ml50 0.1457 1.06×10-6
fl60 4.51×10-5 1.22×10-6 ml60 0.1376 1.06×10-6
TABLE 6
Effective duration and mortality rate of each product (insured)
effective duration
30 40 50 60
fa -3.2065 -2.9768 -2.4200 -1.8099
ma -4.1073 -3.7589 -2.9648 -2.1000
fl 11.4160 10.4906 8.3051 6.6572
ml 13.3312 10.7881 8.0193 5.8648
mortality rate
30 40 50 60
fa 0.000116 0.000168 0.000424 0.001306
ma 0.000121 0.000436 0.001125 0.003044
fl 0.000168 0.000382 0.001155 0.002655
ml 0.000261 0.001 0.002606 0.006274
25
TABLE 7-1 female annuity due TABLE 7-2 female deferred annuity
Comparison of deferred effect of female annuity (θ = 0)
1 unit fa60 1 unit fa30
N f(N) N f(N)
ml30 0.1226 1.56×10-7 ml30 0.0140 2.04×10-9
ml40 0.0863 1.62×10-7 ml40 0.0097 2.04×10-9
ml50 0.0407 1.62×10-7 ml50 0.0046 2.04×10-9
ml60 0.0355 1.62×10-7 ml60 0.0040 2.04×10-9
TABLE 8-1 male annuity due TABLE 8-2 male deferred annuity
Comparison of deferred effect of male annuity (θ = 0)
1 unit ma60 1 unit ma30
N f(N) N f(N)
ml30 0.4903 1.07×10-6 ml30 0.0438 8.54×10-9
ml40 0.338 1.07×10-6 ml40 0.0302 8.54×10-9
ml50 0.1595 1.07×10-6 ml50 0.0142 8.54×10-9
ml60 0.1393 1.07×10-6 ml60 0.0124 8.54×10-9
TABLE 9
Optimal units of life insurances, given 1 unit of annuity (θ = 1)
1 unit fa60 1 unit ma60
fl N ml N fl N ml N
fl30 0.0985 ml30 0.1257 fl30 0.173 ml30 0.2207
fl40 0.1519 ml40 0.1876 fl40 0.2668 ml40 0.3294
fl50 0.1987 ml50 0.2735 fl50 0.3488 ml50 0.4802
fl60 0.3293 ml60 0.4236 fl60 0.5782 ml60 0.7438
TABLE 10
Comparison of gender effect of annuity (θ = 1)
1 unit fa30 1 unit ma30
fl N ml N fl N ml N
fl30 0.3054 ml30 0.3895 fl30 0.5635 ml30 0.7188
fl40 0.4709 ml40 0.5813 fl40 0.869 ml40 1.0728
26
fl50 0.6157 ml50 0.8476 fl50 1.1363 ml50 1.5642
fl60 1.0205 ml60 1.3129 fl60 1.8833 ml60 2.423
TABLE 11-
Optimal units of female life insurance to hedge one unit of annuity60 for both male
and female under different value of θ
1 unit fa60 1 unit ma60
θ fl30 fl40 fl50 fl60 θ fl30 fl40 fl50 fl60
0 4.51×10-5 4.51×10-5 4.51×10-5 4.51×10-5 0 4.51×10-5 4.51×10-5 4.51×10-5 4.51×10-5
0.0001 0.0979 0.0628 0.0096 4.51×10-5 0.0001 0.1413 0.0773 4.51×10-5 4.51×10-5
0.001 0.0985 0.1362 0.1058 0.0825 0.001 0.1696 0.2333 0.1721 0.1244
0.005 0.0985 0.1486 0.1703 0.2132 0.005 0.1723 0.2596 0.2948 0.3647
0.01 0.0985 0.1502 0.1835 0.2597 0.01 0.1727 0.2632 0.3200 0.4502
0.1 0.0985 0.1518 0.1972 0.3215 0.1 0.1730 0.2664 0.3460 0.5639
1 0.0985 0.1519 0.1987 0.3293 1 0.1730 0.2668 0.3488 0.5782
TABLE 12
Optimal units of male life insurance to hedge one unit of annuity60 for both male and
female under different value of θ
1 unit fa60 1 unit ma60
θ ml30 ml40 ml50 ml60 θ ml30 ml40 ml50 ml60
0 0.1226 0.0863 0.0407 0.0355 0 0.4903 0.338 0.1595 0.1393
0.0001 0.1249 0.1269 0.0560 0.0440 0.0001 0.2857 0.3345 0.1805 0.1525
0.001 0.1256 0.1744 0.1368 0.1064 0.001 0.229 0.3305 0.2919 0.2497
0.005 0.1257 0.1847 0.2221 0.2407 0.005 0.2224 0.3296 0.4094 0.4589
0.01 0.1257 0.1861 0.2447 0.3044 0.01 0.2215 0.3295 0.4406 0.5581
0.1 0.1257 0.1874 0.2706 0.4086 0.1 0.2207 0.3294 0.4762 0.7204
1 0.1257 0.1876 0.2735 0.4236 1 0.2207 0.3294 0.4802 0.7438
27
FIGURE 1
Parameter Estimates of x , x , tk in LC model
Panel A. Parameter Estimates of x
Panel B. Parameter Estimates of x
28
Panel C. Parameter Estimates of tk
FIGURE 2
Pricing differences of each product between period and cohort bases
FIGURE 3
Optimal units of male and female life insurance to hedge one unit of annuity for
different ages and genders